Neural integrators for decision making: a favorable tradeoff between robustness and sensitivity

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Perceptual decision-making relies on integrating sensory information over time.
• Neural mechanisms for long-timescale integration often require precise parameter tuning.
• A robust integrator mechanism, with a threshold for input registration, offers a potential solution to this precision requirement.

Purpose of the Study:

• To analyze the tradeoff between input sensitivity and parameter robustness in integrator circuits.
• To evaluate the impact of robust integrators on decision-making performance, particularly in the context of sensory evidence accumulation.
• To investigate the implications of robust integration for neural computation in cognitive tasks.

Main Methods:

• Simulations of coupled neural units modeling evidence integration.
• Mathematical analysis using principles of sequential analysis.
• Focus on the random dot motion discrimination task with experimentally constrained stimulus parameters.

Main Results:

• Mistuning feedback in integrator circuits degrades decision performance.
• The robust integrator mechanism effectively limits performance degradation due to parameter mistuning.
• Including a robustness mechanism offers minimal performance penalty even in perfectly tuned circuits.

Conclusions:

• Robust integrators provide a mechanism to overcome the need for precise parameter tuning in neural circuits for evidence integration.
• This mechanism enhances decision-making performance and robustness against neural variability.
• Robust integrators are well-suited for supporting evidence integration in various cognitive tasks.